Executive summary

I estimated how cost-effective seeking farmed fish slaughter commitments from retailers might be in Europe using a Monte Carlo simulation model. I focused on France, Italy, and Spain because of the large scale of consumption of farmed fish species (Gilthead Seabream, European Seabass, and small Rainbow Trout) that don’t benefit from stunning in these countries.
My estimates suggest that, if initial pilot work proves successful, future marginal spending in these countries could potentially benefit between 2 and 36 animals per dollar spent. This is somewhat below estimates of historic chicken campaigns (10 to 280 animals per dollar), but is plausibly comparable to marginal chicken grants once decreases in cost-effectiveness are taken into account. However, the number of animals affected is several magnitudes lower than estimates of ongoing shrimp stunning efforts (880 to 1900 shrimp per dollar per year).
Farmed fish slaughter commitments might only affect between 0.4 and 10 fish hours per dollar. This is several orders of magnitude shorter than estimates for historic chicken campaigns (10 to 120 chicken years per dollar), largely reflecting the very short duration of fish slaughter (5 to 40 minutes for Seabream and Seabass) compared to the lifespan of layer hens and broilers.
Farmed fish slaughter commitments do not look cost-effective using an estimation method that prioritizes duration and does not assign a high moral value to averting intense suffering. Using philosophical assumptions and placeholder estimates from the Rethink Priorities Moral Weights Project, I estimated the cost-effectiveness of farmed fish slaughter commitments to be between $10.4K and $114M per DALY averted. To put these numbers in context, $50 per DALY averted is considered a proxy for some of the most promising human global health and development interventions. The best animal interventions are often considered to be even more competitive than this.
But farmed fish slaughter commitments could seem more competitive in estimation frameworks that prioritize averting the most intense types of suffering. Fish slaughter commitments beat a theoretical funding bar of $50 per DALY in 50% of simulations for someone who was willing to avert 1 year of fish suffering during slaughter over gaining 72 years of human life at full health.
Allocating resources for farmed fish slaughter corporate commitment work could potentially be justified on account of moral and empirical uncertainty around the duration and severity of pain. It could also be justified to better understand the tractability of fish welfare work in these countries, and to help lay the ground for non-slaughter welfare improvements (for example, by getting retailers to recognize that fish are sentient).
Corporate commitments for non-slaughter welfare improvements have the theoretical potential to look more cost-effective than slaughter improvements. But further work needs to be done to establish what the most tractable asks might be. It’s also unclear how competitive such interventions will look once the costs of developing consensus for specific welfare asks and running feasibility pilots is taken into account.

Caveats and limitations

This cost-effectiveness analysis is prospective rather than retrospective. It is aimed at highlighting the theoretical potential of this intervention, but it is not grounded on the historical achievements of a group of NGOs or rigorous efforts to identify causal mechanisms. While efforts were made to ensure the assumptions and modeling approach adopted are reasonable, the prospective and hypothetical nature of this work means it should be considered much weaker evidence than an evaluation supported by, for example, a meta-analysis of randomized-controlled trials (RCTS).

My cost estimates only cover future marginal program spending to secure commitments, and therefore probably underestimate the full underlying costs of fish slaughter corporate commitment work. My estimates of cost-effectiveness focus on marginal spending by NGOs to secure farmed fish slaughter commitments in the relevant countries. But they do not account for historical spending on farmed fish slaughter that may have supported this work (e.g., general fish welfare research, policy efforts, research conducted in other countries). They also do not take into account the costs of additional research to identify and address practical constraints to implementation, or the costs associated with monitoring that retailers and producers comply with their commitments. My estimates therefore most likely overestimate the cost-effectiveness of fish slaughter corporate commitment work – though I don’t think this would affect the main conclusions.

My modeling approach makes short-duration interventions that avert intense pain look less cost-effective. The methodology I used for estimating $/DALYs, which multiplies duration by potential welfare gain, tends to make interventions that address short episodes of very intense pain look less effective than those addressing longer-duration but lower-severity pain. There is, however, moral and empirical uncertainty over whether averting intensity or duration is more important, and a lack of empirical evidence on the duration of intensity of pain that fish might experience during conventional slaughter, and how much this might be averted through humane stunning methods.¹ I attempted to partially address this by also presenting results using an alternative approach (which I call the “moral value approach”), but I appreciate that even this approach might not fully address the concerns of readers who are primarily interested in averting high-intensity pain.

Several of my estimates rely on placeholder estimates for the welfare range of fish developed using a methodology that gives lower scores to less-researched species. In order to produce cost-effectiveness estimates in $/DALY space, I took placeholder estimates of the welfare range of salmon relative to humans from the Rethink Priorities Moral Weights Project and assumed these apply to other fish species. There are a number of thorny methodological issues associated with aggregating the evidence to estimate the size of welfare ranges, and the estimates I used are best regarded as placeholders rather than definitive estimates. Notably, the aggregation methodology treated missing data as evidence against sentience and for larger welfare range differences. The authors suspect the lower welfare ranges for fish relative to pigs and chickens is largely due to a lack of research.

General difficulties in comparing cost-effectiveness of interventions across species – In order to make my results useful for readers, I attempted to compare how the number of animals potentially affected through this fish stunning commitment might compare to chicken and shrimp interventions. I also converted the results into the cost per DALY averted, a ‘common currency’ metric that helps facilitate comparisons with human global health and development interventions. In order to convert the results into $/DALY, I needed to make assumptions about the potential welfare gain arising from the intervention (without empirical evidence to support my choice). I also needed to make assumptions about the percentage improvement in the human welfare range over a given duration that might be equivalent to averting a DALY. My choice of assumptions could be perceived as unfavorable for fish, though I don’t think using alternative assumptions would materially change the conclusions. Attempting to compare cost-effectiveness of interventions across causes and species is an extremely difficult task (see the Rethink Priorities Moral Weight and CURVE sequences).

I do not fully take into account some strategic considerations, e.g., trade-offs between chicken asks and fish welfare asks – While I do give some consideration to ongoing chicken welfare work when considering the size of retailers that NGOs might engage in different scenarios, my analysis does not give full consideration to the possible trade-offs between fish welfare and chicken welfare objectives. Attempting to engage retailers on improving fish welfare could be a distraction from efforts to secure and implement chicken welfare commitments. It is unclear whether the primary bottleneck for chicken work is funding or talent, or the extent to which talent pools for chicken advocacy and fish advocacy overlap.

Generalizability to other species and countries – My analysis is focused on securing stunning commitments for three farmed fish species (European Seabass, Gilthead Seabream, and the small variety of Rainbow Trout) through grocers in three countries (France, Italy, and Spain). While I selected the scope because I perceived there to be high potential and this might reasonably affect priors about how promising other farmed fish interventions look, the extent to which the results can be generalized across to other farmed fish species, geographies, engagement methods, and fish welfare interventions is unclear.

General caveats around cost-effectiveness analyses – Šimčikas (2019) lists a number of ways in which cost-effectiveness estimates can be misleading. Many of these apply to this piece.

Definitions

I use a number of terms repeatedly throughout this report, which are briefly explained here.

I use the term fish to refer to finfish rather than shellfish, such as crustaceans and mollusks. I use Seabass as shorthand to refer to European Seabass (Dicentrarchus labrax), Seabream to refer to Gilthead Seabream (Sparus aurata), and Trout to refer to Rainbow Trout (Oncorhynchus mykiss).

Rainbow Trout is harvested at lots of different sizes. I use the terms small or portion to refer to any fish of this species harvested at a weight below 1.2kg, with those harvested at a greater weight described as large. My quantitative analysis focuses on the small and portion varieties. Any reference to this species from the methods section or beyond will refer to the small or portion varieties unless otherwise specified.

In the methods section, I typically define input assumptions using the upper and lower values of a 90% subjective confidence interval. In short, this means I’d assign a 90% probability to the true value being between the lower and upper values, a 5% probability to the true value being above the upper value, and 5% probability to the true value being below the lower value. In the results section, I typically present results as a 90-percentile range based on 5th and 95th percentile values of the Monte Carlo simulations.

Throughout the report, I use $/DALY or cost per DALY as shorthand for the cost of averting a DALY, where a DALY is a disability-adjusted life year and represents the loss of the equivalent of one year of human life at full health. The cost per DALY averted is a ‘common currency’ metric that several organizations use to compare interventions.

When presenting numbers, I use K to represent a thousand, M for a million, B for a billion (10^9).

Introduction and motivation

Farmed fish are numerous and neglected. Advocates are now pursuing a range of interventions to improve farmed fish welfare. But little is known about the potential cost-effectiveness of these interventions.

Fish are the vertebrates most numerously farmed and killed for food production. Mood et al. (2023) estimate that between 78 billion and 191 billion farmed finfish were killed for food in 2019, compared to around 80 billion birds and mammals. Šimčikas (2020) estimated that around 103 billion finfish farmed for food might be alive at any time, compared to 2018 UN Food & Agriculture Organisation estimates of 31 billion birds and mammals.

Despite the scale of farming, fish welfare has been historically neglected by the animal advocacy movement.² However, resources allocated towards farmed fish welfare have increased in recent years,³ with the creation of new animal advocacy organizations specifically dedicated to farmed fish welfare,⁴ and a growing number of organizations including fish and other aquatic animals into their remit in some way.⁵

Advocates are now pursuing a number of interventions to improve farmed fish welfare. These include supporting fish welfare research, raising awareness through undercover investigations, influencing seafood certifiers, policy engagement, and corporate engagement – including producers, suppliers, and retailers.

There have been relatively few independent efforts to estimate the cost-effectiveness of efforts to improve farmed fish welfare. Some organizations now report their own estimates of the number of fish that they may have helped per dollar spent (e.g., Fish Welfare Initiative’s impact estimates, Aquatic Life Institute’s own estimates of the number of fish benefiting from its certifier engagement program). Animal Charity Evaluators scores cost-effectiveness of the programs of the charities it evaluates, some of which include fish welfare programs.⁶ The Rethink Priorities cross-cause cost-effectiveness model also includes a “generic carp intervention,” but is not intended to establish definitive conclusions about the relative effectiveness of projects.

Why evaluate fish slaughter commitments in France, Italy, and Spain

I focused on evaluating the cost-effectiveness of seeking farmed fish slaughter commitments from retailers in France, Italy, and Spain. This is because of the historic success of corporate commitment work for chickens in Europe, the scale of consumption of farmed fish species in these countries that don’t benefit from stunning before slaughter, and NGO interest in pursuing this work.

Corporate commitment work for chickens is widely recognized as a cost-effective farm animal welfare intervention. Bollard (2016) identified the cost-effectiveness of cage-free campaigns, in terms of animal suffering per dollar, to be better than any alternatives he was aware of at the time. Capriati (2018) reported the expected benefits for corporate campaigns to be extremely large. Šimčikas (2019) estimated that historic chicken corporate campaigns affect 9 to 120 years of chicken life per dollar spent between 2005 and 2018. This type of intervention is still considered a top option even after taking into account declining returns in recent years.

Europe is ahead of other regions when it comes to chicken welfare progress – Europe has secured more corporate commitments for chickens than any other continent.⁷ A majority of the major grocers in France and the Netherlands signed up to the European Chicken Commitment,⁸ and caged housing systems for layer hens are banned or due to be banned in several European countries.⁹ The progress made on chickens may mean that food-service businesses in Europe may be amenable to making welfare commitments for other neglected species, and animal advocacy groups might be in a better position to prioritize fish welfare work than in other regions.

There is evidence that grocers in Europe might be willing to make fish welfare commitments – Several major German grocers are signed up to the Aquaculture Welfare Standards Initiative and have made public commitments to sell only third-party certified farmed fish.¹⁰ A number of UK grocers have animal welfare policies that require their own brands of farmed finfish to be stunned before slaughter.¹¹ Aquaculture producers in Turkey report demand from their customers in the UK and Netherlands as the most important incentive for using electrical stunning systems.

Corporate commitment work could have some advantages over and/or be complementary to fish policy work – At the time this project started, it seemed plausible that the EU might include slaughter reforms for farmed fish as part of a package of broader animal welfare reforms. The proposals featured in a draft EU Commission impact assessment document leaked in June 2023 suggested farmed fish slaughter reforms would be limited to EU production and have a 10-year implementation period for Gilthead Seabream and European Seabass. Corporate commitments have the potential to be implemented faster.¹² They can also apply to extra-EU imports, which is important given that a majority of EU farmed fish for consumption (by weight) is produced outside the EU.¹³ At the time of writing, progress on EU fish slaughter reforms seems extremely unlikely before the terms of the current European Parliament and Commission expire. Progress through corporate commitments could nevertheless widen the support base when opportunities for policy reform arise in the future.

When measured by the number of individuals slaughtered, EU farmed fish consumption in Europe is dominated by Gilthead Seabream, small Rainbow Trout, European Seabass, and Atlantic Salmon – Figure 1 below provides estimates of the number of finfish slaughtered (by species group) to meet EU27 consumption in 2021. Using the “mid-point estimates,” four species account for around three quarters of consumption (by number of individuals) where specific species information is available. Based on mid-point estimates, around 391m Gilthead Seabream were slaughtered to meet EU farmed fish consumption in 2021, compared to 245m small Rainbow Trout, 237m European Seabass, and 194m Atlantic Salmon.

Figure 1: Estimated individuals slaughtered to meet EU2027 farmed fish consumption in 2021

Source: Rethink Priorities analysis of EUMOFA apparent consumption estimates. Fish weight data obtained from multiple sources.¹⁴

Notes: Solid bars represent a central estimate of the number of fish slaughtered each year, typically using the mean of the upper and lower estimated mean weight. Error bars indicate range of possible values if mean slaughter weight was at the upper or lower estimated mean weight. Estimates do not include fish that die on farms before slaughter.

A majority of Gilthead Seabream, European Seabass, and Rainbow Trout are not stunned before slaughter – Figure 2 below provides estimates of the share of EU farms that stun fish before slaughter, based on a draft EU Commission impact assessment document leaked in June 2023. The report suggested less than 5% of Seabream and Seabass farmed in the EU are stunned before slaughter, with the majority killed by immersion in ice slurry, followed by gill cutting. A recent survey of Turkish producers accounting for around three quarters of Turkish Seabass and Seabream production suggested around 40% of their production is electrically stunned before slaughter (Kıraç Uncu & Özen, 2023). Turkey is the country with the largest global production of these species and a major exporter to the EU. EUMOFA balance sheet data suggest extra-EU imports have accounted for around 20-30% of consumption of these species in recent years. The leaked EU Commission report suggested between 20% and 50% of EU trout farms stun their fish. Where Rainbow Trout are not stunned, common slaughter methods include asphyxia in ice and immersion in CO₂-saturated water (with or without live chilling). The leaked EU Commission report suggested that over 90% of farms producing salmon in the EU stun fish before slaughter, and it is likely that a similar proportion of imported farmed salmon are stunned too.¹⁵

Figure 2: Share of EU farms that stun fish before slaughter

Source: Draft EU Impact Assessment document leaked in June 2023.

Notes: For Atlantic Salmon and Rainbow Trout, the EU Impact Assessment document refers to the share of farms stunning fish, whereas for carp and seabream/seabass, the document appears to refer to the share of fish.

Conventional slaughter methods are not considered acceptable for Seabream, Seabass, and Rainbow Trout – The World Organization for Animal Health (WOAH, formerly known as OIE) recommends that “as a general principle, farmed fish should be stunned before killing.” It assesses chilling fish in ice slurry, and immersion in CO₂-holding water as associated with poor welfare. Species-specific scientific opinions by the European Food Standards Agency (EFSA) Panel on Animal Health and Welfare also identified all commercially used methods of slaughter, except electrical and percussive stunning for Rainbow Trout, to be associated with poor welfare.

Consumption of seabream, seabass, and small rainbow trout is dominated by Italy (231M individual fish), Spain (126M fish), and France (99m fish) – Figure 3 below provides estimates of the number of farmed small Rainbow Trout, Gilthead Seabream, and European Seabass that are slaughtered each year to support consumption by different EU27 countries, based on EUMOFA estimates of apparent consumption and mid-point estimates of slaughter weight.

Figure 3: Annual number of Seabass, Seabream, and small Rainbow Trout slaughtered to support consumption in EU27 countries

Source: Rethink Priorities analysis of EUMOFA apparent consumption estimates of European Seabass (2016), Gilthead Seabream (2019), and small Rainbow Trout (2020).¹⁶

There are NGOs interested in pursuing fish slaughter corporate commitment work for these species – At the outset of this project, I interviewed 23 different stakeholders on a range of issues related to fish welfare, including appetite to pursue corporate commitment work to improve farmed fish welfare. A number of NGOs operating in Mediterranean countries indicated an interest in pursuing this work, and had formed an informal working group to facilitate collaboration. These NGOs include:

Essere Animali – who operate in Italy, have a corporate fish ask template that includes slaughter as an essential element, and have commissioned work evaluating humane slaughter methods for farmed fish in Italy and Greece;
Animal Welfare Observatory (AWO, formerly known as Equalia) – who operate in Spain, have a fish stunning corporate ask, ¹⁷ and have worked with APROMAR, the Spanish aquaculture producer association, to help develop welfare guidelines for Seabass and Seabream;
Welfarm – who operate in France and have published recommendations on fish slaughter with a stated aim of lobbying retailer “purchasing offices about their supply policies” (pp.2);
Compassion in World Farming – who operate in France, Italy, and Spain, have a food business team that includes Fish Welfare Specialists, and worked with AWO on the development of welfare guidelines for Seabass and Seabream in Spain; and
Future for Fish, who are engaging producers in Turkey on electrical stunning systems for Seabream and Seabass and have secured stunning commitments from Turkish producers and retailers.

Methodology

I estimated the cost-effectiveness of this intervention using a Monte Carlo simulation model in R with 10,000 simulations (full code available on GitHub). Using assumptions about how successful the intervention might be, I first estimated the potential number of farmed fish and fish hours affected per dollar spent. Using further assumptions about the potential reduction in suffering from the intervention, and assumptions to facilitate interspecies comparison, I then calculated the results as a $/DALY range and the share of simulations outperforming three different $/DALY benchmarks.

Estimating the number of unstunned fish consumed annually and slaughter duration by country

To estimate the number of fish who might be affected by slaughter commitments, I first estimated the number of farmed fish consumed annually in each country that do not currently benefit from stunning before slaughter, and the aggregate duration of slaughter (under conventional methods) for these fish. Figures 4 and 5 below describe how these were calculated, Table 1 below provides a summary of the data and assumptions used in these calculations, and Annex 1 provides further detail on the specific numbers used.

Figure 4: Estimating number of unstunned farmed fish consumed annually and aggregate slaughter duration

Figure 5: Estimating annual consumption of individual farmed species

Table 1: Estimating the number of fish affected per dollar

Term	Description
Annual consumption of farmed species	Estimated number of individual farmed European Seabass, Gilthead Seabream, and small Rainbow Trout slaughtered in order to support consumption in France, Italy, and Spain.
Apparent consumption	Estimated consumption (in live weight equivalent) of selected species by country provided by the European Market Observatory for Fisheries and Aquaculture Products (EUMOFA) in various case study reports. Measured in metric tonnes. EUMOFA estimates are based on production plus imports less exports, with adjustments made to convert trade data related to weights of processed fish products into live weight equivalent.
Estimated mean slaughter weight	Plausible range of the mean slaughter weight for each species based on Mood and Brooke (2019a).
Share not stunned	Estimate of the share of each farmed fish species that are not stunned via an acceptable method (electrical stunning or percussive stunning). Ice slurry is sometimes called thermal stunning, and immersion in CO₂saturated water is sometimes called CO₂stunning. Neither are considered an acceptable slaughter method unless fish are stunned beforehand either by electrical or percussive means.
Slaughter duration	Typical time to unconsciousness (or death) for each species under conventional slaughter methods.

Estimating the number of fish affected and fish hours affected per dollar

To estimate the number of fish affected and fish hours affected per dollar, I developed a set of scenarios for each country based on the market share of grocers that might make commitments, conditional on a marginal amount of program spending on fish slaughter commitments. I multiplied the number of unstunned fish consumed annually (and aggregate slaughter duration of those fish) in each country by the market share of grocers making commitments and the share of fish sales going through grocers. I then multiplied by an assumption regarding years of impact and applied an implementation discount, before dividing by the assumed costs of securing commitments in the scenario for each country.

Figure 6 visually illustrates the calculation method, Table 2 provides a high-level description of each of the terms, and Tables 3 to 5 provide further detail on the specific assumptions for each country.

Note that to estimate the aggregate number of fish affected across multiple countries, I summed the total number of fish (or fish hours) affected across all countries before dividing by the combined cost across all countries.

Figure 6: Estimating the number of fish (or fish hours) affected per dollar

Note: Unstunned farmed fish consumed annually can be replaced with aggregate slaughter duration for those fish to estimate the number of fish hours affected per unit spent.

Table 2 – Definitions of terms to estimate number of fish (or fish hours) affected per dollar

Term	Description
Market share of grocers making commitments	Share of the domestic grocery market occupied by grocers from whom advocates secure commitments, conditional on an amount of marginal program spending in a given scenario.
Probability of success	Probability that grocers make a commitment, conditional on program spending in a scenario.
Costs of securing commitments	Assumption about the marginal spending by animal advocacy group programs to secure slaughter commitments in a given scenario. I assumed spending in Euros and converted to US Dollars at an exchange rate of 1.09 (spot estimate as of January 24, 2024).
Share of fish sales through grocers	Assumption about the share of farmed fish species consumed sold through grocers.
Implementation discount	Percentage discount applied to annual number of fish sold via grocers, to take into account fish that might not counterfactually benefit from stunning as a consequence of a commitment. This could be because a grocer fails to follow through on its commitments or applies the commitment to only a fraction of the fish it sells (e.g., own brand products only), or because a higher-than-average share of fish sold by grocers making commitments are already stunned.
Years of impact	Assumption regarding the number of years of counterfactual impact a grocer stunning commitment might have. Can be thought of as the number of years program spending has brought forward stunning commitments by grocers, relative to a counterfactual without the spending.

Commitment scenarios - market share and success probabilities

I produced two different scenarios for each country: a pilot scenario and a scale scenario.

The pilot scenario is designed to test the receptiveness of grocers in that country to making fish slaughter commitments and to demonstrate proof of concept. This is aimed at illustrating the potential effect of securing fish slaughter commitments from one to two smaller or regional grocers in a country. Smaller grocers may be a suitable choice for pilot work, as they may be more amenable to making a fish slaughter commitment (for example, because of a track record of commitments for other species) or because seeking commitments from such grocers may be less likely to cause a conflict with ongoing work for other species (for example, getting the largest grocers to sign-up to or implement the European Chicken Commitment).

Conditional on a successful pilot scenario, the scale scenario is designed to illustrate the potential impact of a bigger program of work to secure commitments from one to two of the larger (i.e. top 10) grocers in each country.

Tables 3 and 4 describe the specific assumptions for the pilot and scale scenarios respectively.

I developed the assumptions for both scenarios after examining data on the market share of grocers in each country and examining data on which grocers had made cage-free and broiler reform commitments for chickens.¹⁸ I sought and received feedback on the specific assumptions for Spain and Italy from AWO and Essere Animali respectively, and attempted to broadly align assumptions for France with those for Spain and Italy. Market shares and success probabilities were modeled using a beta distribution such that 90% of observations would fall between the lower and upper values reported in the tables.

I used the same probability of success and marginal program costs in each country for simplicity, though in reality it is likely to be different across each country. I assumed uncertainty only in the effects and not the costs to simplify the interpretation of results, depending on whether they are presented in impact per dollar spent, or cost per unit of impact.¹⁹

Table 3 – Pilot scenario assumptions

Country	Market share of grocers		Probability of success		Marginal program costs in each country (Euros)
	Lower	Upper	Lower	Upper
France	0.1%	1%	30%	60%	100K
Italy	0.1%	2%
Spain	0.1%	1%

Table 4 – Scale scenario assumptions

Country	Market share of grocers²⁰		Probability of success		Marginal program costs in each country (Euros)
	Lower	Upper	Lower	Upper
France	1%	24%	15%	35%	250K
Italy	4%	14%
Spain	3%	10%

Other assumptions – years of impact, implementation discounts, and grocery sales share

Table 5 below describes the assumptions I used for years of impact, implementation discount, and the share of fish sales going through grocers. For simplicity, I used the same assumption for all countries.

Table 5 - General model assumptions

Assumption	Range		Statistical distribution
Assumption	Lower	Upper	Statistical distribution
Years of impact	5 years	20 years	Gamma
Implementation discount	5%	45%	Beta
Share of fish sales through grocery sector	60%	75%	Beta

The final results are highly sensitive to the years of impact assumption,²¹ but it is an assumption that is very difficult to calibrate, and I have relatively low confidence in my assumption here. I used 5 years for the lower end of my subjective confidence interval. This was based on an assumption that it might take around 5 years before grocers have implemented commitments (around a year or two to secure commitments, three to four years for grocers to implement the commitments),²² and an assumption that the farmed fish they sell may be stunned in 10 years time – either as a result of EU-wide policy²³ or environmental groups pushing third-party certification schemes that have fish stunning requirements. I based the upper end of my subjective interval range on the mid-point of 90% subjective confidence interval bounds used by Šimčikas (2019) in his cost-effectiveness estimate for chickens (4 to 36 years).

I found relatively little information to base my implementation discount assumption on, and also have relatively low confidence in my assumption here. My assumptions were informed by the subjective confidence intervals used by Šimčikas (2019), which assumed follow-through rates of 48% to 84% for cage-free campaigns, and 1% to 94% for broiler welfare commitment.²⁴ My intuitions are that the appropriate implementation discount should reduce the number of animals affected (or duration) to a greater extent than the follow-through rate used by Šimčikas (2019). This is for two reasons. First, I expect supply chain complexities associated with the international trade of fish will make it more difficult for retailers to ensure all fish sold complies with a commitment in all circumstances. Second, my implementation discount is also intended to adjust for fish sold by retailers who already stun a greater proportion of stunned fish than the EU average, whereas Šimčikas (2019) adjusts for this elsewhere in his model.

The share of fish sales through the grocery sector assumption was calibrated using data from the 2022 EUMOFA Fish Market Report, which suggests that between 80% and 83% of unprocessed fishery and aquaculture products (by volume) are sold through retail channels in the countries of interest.²⁵ I then adjusted these figures down by 10% to 25% to take into account how some retail fish sales might not go through grocers.²⁶

Converting the results into $/DALY averted equivalents

To enable this intervention to be compared against non-animal interventions, I estimated the impact of the fish slaughter commitments in $/DALY space – a ‘common currency’ unit used by many Effective Altruist organizations. To do so, it was necessary to make a number of additional assumptions. These include philosophical assumptions (of which there are many arguments for and against), assumptions about the magnitude of the potential welfare gain arising through the stunning intervention (for which we don’t have empirical data), the size of fish welfare ranges relative to humans (which is methodologically challenging to estimate), and an assumption about how to define averting a DALY relative to the size of the human welfare range.

Recognising the uncertainty associated with these assumptions, I present the results using three different approaches. The first is a direct $/DALY range. This is arguably the most straightforward to articulate, but also requires the most assumptions to estimate. The second and third approaches calculate the share of simulations beating three selected $/DALY benchmarks. The second approach, which I call the “$/DALY benchmark welfare range” approach, follows a similar method to that of the direct $/DALY range, but relaxes the need for specific assumptions about the welfare gain associated with the intervention. The third approach, which I call the “$/DALY benchmark moral value” approach, relaxes the need for specific assumptions about the welfare range of fish relative to humans, and allows for worldviews that assign greater (or lower) moral value to averting suffering through the slaughter intervention relative to increasing the healthy lifespan for humans.

To calculate the direct $/DALY range, I assumed that the stunning intervention leads to a welfare improvement between 10% and 50% of the entire fish welfare range (both positive and negative), that lasts for 50% to 90% of the duration of slaughter without stunning. I do not have empirical data to justify these assumptions, but I believe them to be reasonable best guesses.

To convert welfare gain from the intervention into human welfare range space, I assumed that the welfare range for all fish species of interest, relative to humans, is the same as for salmon.²⁷ I used the probability-of-sentience and rate-of-subjective-experience-adjusted welfare ranges for salmon calculated as part of the Rethink Priorities Moral Weights Project published here. I incorporated this assumption into the model using a beta distribution with a mean of 0.13 and a standard deviation of 0.17.

To convert this into DALYs averted, I assumed that an intervention that lasts for a year and generates an average welfare improvement of 50% of the entire human welfare range is equivalent to averting a DALY.²⁸ Numerically, this approach generates identical results to those used in the illustrative example in the Introduction to the Moral Weight Project. But it is less restrictive in that it does not require the specific assumptions used in that example, namely valence symmetry and the assumption that the relevant best human welfare state is equal to the average welfare of the average human in full health.

Under the $/DALY benchmark welfare range approach, I follow the same assumptions as above, but don’t make a specific assumption about the welfare gain associated with the stunning intervention. Instead, I estimate the share of simulations beating three different benchmarks for a range of different hypothetical values for the welfare gain of the intervention, all expressed as a % of the entire fish welfare range.

Under the $/DALY benchmark moral value approach, I estimate the share of simulations beating selected $/DALY benchmarks. These are presented conditional on hypothetical assumptions about the subjective moral value of averting 1 year of fish suffering through the intervention relative to gaining 1 year of human life at full health.

The three different cost-effectiveness benchmarks are $50/DALY, $1K/DALY, and $70K/DALY. $50/DALY is the rough funding bar for global health and wellbeing interventions by Open Philanthropy, and is intended to serve as a proxy for the cost-effectiveness of some of the most promising human global health and development interventions. The other benchmarks are based on estimates of the average cost per DALY averted from health expenditure in low ($1K/DALY) and very high ($70K/DALY) human development index (HDI) countries in Daroudi et al. (2021).

Figure 7 illustrates the general method for estimating $/DALY under all these approaches. Table 6 summarizes the philosophical assumptions that underpin all the approaches. Table 7 provides a description of the different approaches and any additional approach-specific assumptions. Figure 8 illustrates how the DALY conversion factor was calculated in each of the approaches. Note that in Table 7 and Figure 8, the approaches appear in increasing order of the number of assumptions required, rather than the order in which they appear in the results section.

Figure 7: General method for estimating $/DALY

Table 6: Assumptions underpinning $/DALY conversions across all approaches

Assumption	Description
Utilitarianism	We ought to maximize (expected) utility.
Hedonism	Welfare is determined wholly by positively and negatively valenced experiences.

Source: Adopted from Fischer (2022)

Table 7: Overview of approaches and additional assumptions needed

Outcome measure	Approach	Description	Additional assumptions over and above those described in Table 6
Share of simulations beating $/DALY benchmark	Moral value approach	Share of simulations more cost-effective than selected $/DALY benchmarks, conditional on hypothetical values of how morally valuable averting 1 year of fish suffering through the intervention is relative to averting one DALY.	n/a
Share of simulations beating $/DALY benchmark	Welfare range approach	Share of simulations more cost-effective than selected $/DALY benchmarks, conditional on the welfare gain for fish arising from the intervention, expressed as a share of the entire fish welfare range (both positive and negative).	Unitarianism – equal amounts of welfare count equally, regardless of whose welfare it is. Improving human welfare by 50% of the entire human welfare range (positive and negative) for 1 year is equivalent to averting one DALY. Welfare ranges for all affected species (relative to humans) are equivalent to those for salmon.
Direct $/DALY range		$/DALY range	As above, plus: Intervention results in a gain of welfare of 10% to 50% of the entire (positive and negative) welfare range of the fish for 50% to 90% of the duration of conventional slaughter.

Figure 8: DALY conversion factors across different approaches

Notes: Boxes highlighted in yellow indicate an input where I do not assign a specific value, but present how changing this value affects the share of simulations beating the benchmark.

Results

Number of fish affected

Using the methodology and assumptions set out above, my model estimates that work to secure slaughter stunning commitments from retailers in France, Italy, and Spain could benefit 1 to 14 fish per dollar spent in the pilot scenario, and 2 to 36 fish per dollar spent in the scale scenario.

Figure 9 provides a density plot illustrating the distribution of the number of fish affected per dollar for each scenario. Note that the x-axis is log-scale. If viewed in a linear scale, the distribution plot would appear right-skewed, with much of the distribution lying among lower values, and a long right-tail.

Figure 9: Number of fish affected per dollar spent – density plot

Table 8 compares these results to those for chicken corporate campaigns (from Šimčikas, 2019). The potential number of animals affected per dollar for fish stunning corporate commitments, even in the scale scenario (2 to 36 animals per dollar), is lower than historical estimates for chicken corporate commitments (10 to 280 animals per dollar) published by Šimčikas (2019). But the estimates are within an order of magnitude, and there is some overlap in the confidence intervals.

Furthermore, there are reasons to think the cost-effectiveness of current marginal chicken campaign grants might be significantly lower than historical ones.²⁹ If this is the case, it’s plausible that the number of fish affected per dollar in the scale scenario could be comparable to that of marginal chicken grants.³⁰

Table 8: Number of animals affected per dollar spent – comparison with chicken campaigns

Source	Intervention	Scenario	Statistic
			Mean	Percentile
				5th	95th
Shah (2024)	Fish stunning corporate commitments	Pilot	4.6	0.6	13.5
		Scale	13.3	1.9	36
Šimčikas (2019)	Chicken corporate campaigns	Cage-free and broiler (combined)	64	10	280

Table 9 compares the results to shrimp stunning efforts (from the Shrimp Welfare Project, 2023). Note that in order to allow comparisons with the shrimp stunning efforts which are reported in shrimp per dollar per year, I re-calculated the fish stunning results assuming only a single year of impact, rather than my baseline assumption of 5 to 20 years.

However, the number of animals affected per dollar per year for fish stunning commitments is several orders of magnitude lower than the Shrimp Welfare Project’s own estimates of its shrimp stunning corporate partnership work. Fish corporate commitments are estimated to affect 0.2 to 2.8 fish per dollar per year in the scale scenario, compared to estimates of 880 to 1900 shrimp per dollar per year.

Table 9: Number of animals affected per dollar per year – comparison with shrimp stunning efforts

Source	Intervention	Scenario	Statistic
			Mean	Percentile
				5th	95th
Shah (2024)	Fish stunning corporate commitments	Pilot	0.4	0.1	1.1
		Scale	1.2	0.2	2.8
Shrimp Welfare Project (2023)	Shrimp stunning corporate partnerships		1400	880	1900

Note: Fish stunning corporate commitments calculated assuming only a single year of impact in order to facilitate comparisons with the Shrimp Welfare Project estimates, which are presented as animals per dollar per year.

Number of fish hours affected

My model estimates fish corporate commitments to affect 0.1 to 3.8 hours of fish life per dollar spent in the pilot scenario, and 0.4 to 10.4 hours per dollar in the scale scenario. Figure 10 provides a density plot illustrating the distribution (note log-scale), and Table 10 compares these results to those of historical chicken corporate campaigns (from Šimčikas, 2019).

The number of animal hours affected per dollar spent is estimated to be around 5 orders of magnitude lower for the fish stunning corporate commitment scale scenario than for historic chicken campaigns (0.1M to 1.1M chicken hours or 9.6 to 120 chicken years).

The large scale of the difference is mainly due to the relatively short duration of fish slaughter. Seabass and Seabream may remain conscious for 5 to 40 minutes during slaughter via ice slurry. By comparison, Šimčikas (2019) assumed chickens affected by broiler campaigns have an average lifespan of 43 to 48 days and layer hens affected by cage-free campaigns have an average lifespan of 1.1 to 1.5 years.

Even if it is assumed that the cost-effectiveness of marginal chicken campaign grants is several times lower than historic grants, it is likely that the number of hours of animal life affected per dollar will remain several orders of magnitude lower for fish stunning corporate commitments than chicken campaigns.

Figure 10: Hours of fish life affected per dollar spent – density plot

Table 10: Number of animals hours affected per dollar spent – comparison with chicken campaigns

Source	Intervention	Scenario	Statistic
			Mean	Percentile
				5th	95th
Shah (2024)	Fish stunning corporate commitments	Pilot	1.2	0.1	3.8
		Scale	3.6	0.4	10.4
Šimčikas (2019)	Chicken corporate campaigns	Cage-free and broiler (combined)	0.4M	0.1M	1.1M

Note: Šimčikas (2019) presents results of historic chicken campaigns as years of chicken life per dollar. I converted this to hours.

I did not produce comparable estimates for the number of animal hours affected by ongoing shrimp stunning efforts. But so long as the duration of shrimp slaughter is within 2 orders of magnitude of the duration of slaughter for affected fish species, it is very likely that the number of animal hours affected per dollar will remain several orders of magnitude lower for fish stunning interventions than for shrimp stunning interventions, mainly because the shrimp stunning interventions appear to affect so many more animals per dollar.

Cost-effectiveness in $ per DALY averted

I now turn to describing the results of the fish stunning intervention estimated in $/DALY space. To help contextualize the findings, I compare the results against three benchmarks which aim to serve as proxies to illustrate the potential cost-effectiveness of some of the most promising human global health and development interventions ($50/DALY), general health expenditure in low human development index (HDI) countries ($1K/DALY) and health expenditure in very high HDI countries ($70K).

I chose to compare estimates for the fish stunning intervention against human development benchmarks because, based on the number of animal hours affected, it seemed extremely unlikely that the fish stunning intervention would look more cost-effective than chicken campaigns when converted into $DALY/space and estimated using a comparable method. Given that chicken campaigns are often considered more cost-effective than many of the most promising human global health and development interventions,³¹ I thought it would be more instructive to explore how the fish stunning intervention might compare to less competitive benchmarks.

Figure 11 and Table 11 below illustrate the distribution of simulations estimating the cost-effectiveness of fish slaughter corporate commitment work expressed in cost per DALY averted. As noted in the methodology section, these results are conditional on assumptions that the size of the fish welfare range (relative to humans) for all affected species is equivalent to that of salmon (previously published by Rethink Priorities); that the stunning intervention provides welfare benefits of 10% to 50% of the entire welfare range of fish for 50% to 90% of the duration of slaughter without stunning; and that an intervention that improves human welfare by 50% of the human welfare range for a year is equivalent to averting a DALY. Figure 11 provides a density plot (log-scale), and Table 11 provides summary statistics of the same data.

Figure 11: Cost-effectiveness in $/DALY – density plot

90% of simulations are estimated to have a cost-effectiveness between $29.9K and $366M per DALY averted in the pilot scenario, and $10.4K to $114M per DALY averted in the scale scenario. Across all simulations, the mean cost-effectiveness is $41.1B per DALY averted in the pilot scenario, and $12.3B per DALY averted in the scale scenario, with the large mean value driven by a small number of simulations with a very high cost per DALY averted.

Table 11: Cost-effectiveness in $/DALY – summary statistics

Statistic	Scenario
	Pilot	Scale


Mean	41.1B	12.3B
p1	11.8K	4.11K
p5	29.9K	10.4K
p10	52.3K	17.8K
p25	151K	51.0K
Median	680K	216K
p75	4.92M	1.61M
p90	61.0M	18.1M
p95	366M	114M
p99	26.9B	9.98B

Calculated in this way, fish stunning corporate commitments do not appear cost-competitive with human global health and development interventions. Not a single simulation (across either scenario) produced a cost-effectiveness estimate below $50 per DALY averted, the rough bar used by Open Philanthropy for funding global health and development interventions. None of the pilot scenario simulations, and only one of 10,000 simulations for the scale scenario produced an estimated cost below $1000/DALY, the estimated cost per DALY averted from health expenditure in low HDI countries in Daroudi et al (2021).

Share of simulations beating $/DALY benchmarks under the welfare range approach

Given the uncertainty in welfare gain arising from the stunning intervention, I calculated the share of simulations beating three different $/DALY benchmarks for a range of different welfare gains assumptions.³² These results are described in Table 12, and illustrated visually in Figure 12.

Even under the unrealistic assumption that a stunning intervention results in an improvement in welfare so big that a fish moves from the worst possible valenced experience to the best possible valenced experience (a gain of 100% of its welfare range) over a time period equal to the entire duration of stunning without slaughter, the stunning intervention beats a $1K/DALY benchmark in only 1.2% of scale scenario simulations.

Table 12: Share of simulations beating $/DALY benchmark – welfare range approach

Welfare gain from intervention (as % of fish welfare range)	Pilot Scenario			Scale Scenario
Welfare gain from intervention (as % of fish welfare range)	$50/DALY	$1K/DALY	$70K/DALY	$50/DALY	$1K/DALY	$70K/DALY
1%	0.0%	0.0%	0.0%	0.0%	0.0%	0.4%
5%	0.0%	0.0%	1.8%	0.0%	0.0%	10.3%
10%	0.0%	0.0%	5.9%	0.0%	0.0%	20.0%
25%	0.0%	0.0%	17.5%	0.0%	0.0%	37.0%
50%	0.0%	0.0%	29.0%	0.0%	0.2%	49.2%
75%	0.0%	0.0%	36.1%	0.0%	0.5%	55.5%
90%	0.0%	0.0%	39.4%	0.0%	0.9%	58.2%
95%	0.0%	0.0%	40.4%	0.0%	1.0%	59.1%
99%	0.0%	0.0%	41.2%	0.0%	1.1%	59.6%
100%	0.0%	0.0%	41.4%	0.0%	1.2%	59.8%

Figure 12: Share of scenarios beating $/DALY benchmark – welfare range approach

In the scale scenario, the intervention beats a benchmark of $70K/DALY in around half of simulations where the intervention is assumed to generate welfare gains of 50% of the entire fish welfare range. An implication is that this intervention could be reasonably considered comparable to the cost-effectiveness of health expenditure in very high HDI countries.

To test for sensitivity of whether my results might be driven by specific modeling choices which could be perceived as unfavorable for fish, I recalculated the results assuming that the welfare range for affected species was twice as large as in my baseline assumptions, and that averting a DALY would be equivalent to increasing human welfare by 10% of the human welfare range (rather than 50%). While these adjustments improved underlying $/DALY scores by approximately an order of magnitude, they did not materially change how the stunning intervention looks relative to the $50/DALY and $1000/DALY benchmarks. See GitHub for the full results of this sensitivity analysis.

Share of simulations beating $/DALY benchmarks under the moral value approach

This section describes the share of simulations beating selected $/DALY benchmarks under the moral value approach. This approach is motivated by uncertainty around the assumptions underpinning the welfare range approach (for example, the size of welfare ranges of fish relative to humans) and the possibility that readers may place more value in averting high-intensity suffering than, say, prolonging healthy lifespan for humans.

Results are expressed conditional on the hypothetical moral values associated with improving a year of fish life through the intervention, relative to averting a DALY. A value of 0.1 would be consistent with being willing to forgo gaining 1 year of average human life at full health in order to avert 10 years of fish suffering through a fish stunning intervention, while a value of 20 would be consistent with forgoing 20 years of human life in order to avert 1 year of fish suffering through the intervention.

Figure 13 illustrates the share of simulations beating selected $/DALY benchmarks under the moral value approach. Tables 13 and 14 present the chart data in tabular format, with Table 13 describing the share of simulations beating the benchmark conditional on hypothetical moral values, and Table 14 describing the minimum moral value associated with a given share of simulations beating the benchmarks.

The results of this method suggest that this intervention could beat a $50/DALY benchmark in 50% of simulations for someone who would be prepared to forgo 72 human years of full health in order to avoid a year of fish suffering during conventional slaughter methods in the scale scenario. For the pilot scenario, the equivalent figure would be 224.

Figure 13: Share of scenarios beating $/DALY benchmark – moral value approach

Table 13: Share of simulations beating $/DALY benchmark – moral value approach

Moral value of improving a fish-life year via intervention (relative to averting a human DALY)	Pilot Scenario			Scale Scenario
	$50/DALY	$1K/DALY	$70K/DALY	$50/DALY	$1K/DALY	$70K/DALY
0.01	0.0%	0.0%	0.1%	0.0%	0.0%	2.9%
0.1	0.0%	0.0%	32.6%	0.0%	0.0%	74.7%
0.25	0.0%	0.0%	66.8%	0.0%	0.1%	93.3%
0.5	0.0%	0.0%	85.3%	0.0%	1.0%	97.8%
0.75	0.0%	0.1%	91.5%	0.0%	3.5%	99.0%
1	0.0%	0.4%	94.7%	0.0%	7.6%	99.6%
5	0.0%	21.1%	99.7%	0.1%	63.4%	100.0%
10	0.0%	46.1%	100.0%	1.0%	84.6%	100.0%
25	0.8%	77.7%	100.0%	12.2%	96.3%	100.0%
50	6.1%	91.1%	100.0%	35.3%	98.1%	100.0%
75	12.2%	95.3%	100.0%	52.1%	99.6%	100.0%
100	21.1%	97.2%	100.0%	63.4%	99.8%	100.0%
500	77.7%	99.9%	100.0%	96.3%	100.0%	100.0%
1000	91.1%	100.0%	100.0%	98.9%	100.0%	100.0%
5000	99.5%	100.0%	100.0%	100.0%	100.0%	100.0%

Table 14: Minimum intervention moral value for given share of simulations to beat $/DALY benchmark

Share of simulations beating benchmark	Pilot Scenario			Scale Scenario
Share of simulations beating benchmark	$50/DALY	$1K/DALY	$70K/DALY	$50/DALY	$1K/DALY	$70K/DALY
0%	0	0	0	0	0	0
1%	26.9	1.3	0	10.0	0.5	0
10%	64.6	3.2	0	22.9	1.1	0
25%	114.8	5.6	0.1	38.9	1.9	0
50%	223.9	11.2	0.2	72.4	3.6	0.1
75%	457.1	22.9	0.3	144.5	7.1	0.1
90%	933.3	46.8	0.7	275.4	13.8	0.2
95%	1.48K	74.1	1	426.6	21.4	0.3
99%	3.47K	173.8	2.5	1.02K	51.3	0.7

Summary and discussion

In this report, I describe the results of Monte Carlo simulation analysis exploring the cost-effectiveness of corporate commitment work to ask grocers in France, Italy, and Spain to commit to only selling farmed finfish that are stunned before slaughter. This work is motivated by the large number of farmed Seabream, Seabass, and Rainbow Trout consumed in these countries that do not currently benefit from stunning before slaughter, and the historical success of chicken corporate campaigns in Europe.

Summary of results

My analysis suggests that it seems plausible that pilot work to test this intervention at smaller retailers could affect between 1 and 14 fish per dollar, and between 2 to 36 fish per dollar after scaling up the programs to engage larger retailers. These estimates are somewhat below those from historic chicken campaigns, but possibly comparable to marginal spending on chicken campaigns once decreases in cost-effectiveness are taken into account. The number of animals affected per dollar however seems several orders of magnitude lower than estimates for ongoing shrimp stunning efforts.

Since fish slaughter for affected species is relatively short in length (5 minutes to 40 minutes in the case of seabream and seabass), I find that securing stunning corporate commitments might only affect between 0.4 and 10 fish hours per dollar. This is several magnitudes lower than historic estimates for chicken campaigns (10 to 120 chicken years per dollar). The same is likely to be true for shrimp stunning efforts, given the large number of animals affected by shrimp stunning efforts.

Given it seemed unlikely that fish stunning commitments would look competitive with some of the most promising animal welfare interventions on duration impacted, I sought to explore how this intervention might compare to benchmarks associated with global human health and development interventions. I did this by estimating the potential cost-effectiveness of fish stunning corporate commitments in $/DALY space - a ‘common currency’ metric that several organizations use to compare interventions. Recognising the methodological challenges with translating the estimates into $/DALY space, I used several different approaches.

The first approach, which I termed the “welfare range” approach, used philosophical assumptions and placeholder estimates from the Rethink Priorities Moral Weights Project, and also adopted practical modeling assumptions taken from example calculations in this publication series. This modeling approach tends to favor interventions that affect longer duration, and assigns equal value to interventions that generate improvements in valenced experience of similar magnitude, irrespective of whether the intervention is averting intense suffering or enhancing neutral or mildly positive experiences.

Under the “welfare range” approach, I found that fish stunning interventions seem unlikely to look cost-competitive with general health spending in low human development index (HDI) countries, let alone the most promising global health and development charities. I explored the potential impact of adjusting some assumptions related to $/DALY conversions that might make the results less unfavorable to fish, but this did not materially change the conclusions.

I then used another approach, which I called the “moral value” approach, which relaxes the need for most of the assumptions used under the welfare range approach. Under this approach, conversions to DALY equivalents are calculated using the hypothetical subjective moral value of improving a year of fish life through the intervention relative to gaining an average year of human life at full health (averting a DALY). Using this method, I found the fish stunning could beat a cost-effectiveness benchmark of $50/DALY (a proxy for the most promising global health and development interventions) in 50% of simulations in the scale scenario for someone who sees equal moral value in averting 1 fish year of suffering through the stunning intervention to averting 72 years of human life at full health. Given this, it seems plausible to me that someone who is primarily interested in averting high-intensity suffering might see fish stunning corporate commitments as having the potential to be comparable with the most promising global health and development interventions.

Theoretical potential for non-slaughter interventions to look more effective

A general finding from this work is that the relatively short duration of conventional fish slaughter methods acts as a constraint on how cost-effective the stunning intervention might look in many frameworks used to compare the effectiveness of different interventions. I therefore explored how cost-effective non-slaughter interventions that have the potential to raise welfare over a greater fraction of a fish’s lifespan might look. Given corporate asks for non-slaughter welfare interventions are at a much earlier stage of development than stunning interventions, this analysis is considerably more speculative. Accordingly, the analysis for non-stunning interventions is housed in an Annex rather than in the main body of this report.

I found that, if longer-duration non-slaughter interventions that improve fish welfare can be identified and commitments can be secured at reasonable cost, there is the potential for such interventions to look considerably more cost-effective than stunning interventions under the “welfare range” methodology. I also found that while it seems unlikely that these interventions will beat a cost-effectiveness benchmark of $50/DALY using my baseline assumptions, it’s possible they look competitive with this benchmark under alternative assumptions I also consider reasonable.

Possible reasons to support fish stunning corporate commitment efforts

Taking all the results together, I see two potential reasons for why readers may wish to support secure fish stunning corporate commitment efforts, despite the results implying they might not look competitive with other promising animal or human interventions.

The first reason is the potential instrumental value of fish stunning corporate commitments in helping secure other fish welfare commitments in the future. Fish stunning commitments are interventions that can be secured right now, and securing such a commitment is likely to require retailers to recognise that fish are sentient beings worthy of respect and care. This recognition, coupled with the experience and learnings from securing stunning commitments, may make it a lot easier to secure commitments for longer-duration welfare improvements, if and when they are ready for deployment. Fish stunning commitments may also be a helpful way of reducing uncertainty about the receptiveness to retailers to fish welfare in these countries, which in turn may be helpful in evaluating whether it is worth pursuing longer-duration welfare improvements.³³

The second reason relates to moral and empirical uncertainty around the relative importance of severity and duration of pain, which could potentially justify supporting interventions that avert high-intensity suffering of a relatively short duration.

The “welfare range” framework I used to estimate the cost-effectiveness (which multiplies duration by potential welfare gain) makes interventions that address short episodes of very intense pain look less effective than those addressing longer-duration but lower severity pain. But there is moral and empirical uncertainty over whether averting intensity or duration is more important.

Rethink Priorities organized an expert workshop in 2023 discussing the trade-offs between severity and duration, and how to prioritize between these two dimensions of pain in the absence of compelling empirical evidence. While there were many competing ideas, attendees mostly agreed with a presumption that long-lasting harms were more important, but also recommended placing some credence (~20%) in the view that the most severe harms could cause as much disutility as the longest-lasting harms.

I believe the cost-effectiveness results using the “moral weight” approach, coupled with a desire to allocate at least some of the animal advocacy movement’s resources towards severe harms,³⁴ could potentially justify support for stunning corporate commitment efforts in these countries, despite the headline results in $/DALY not looking particularly competitive.

That said, the same arguments could potentially present an even stronger argument for supporting shrimp stunning initiatives, given the significantly larger number of animals this intervention might affect. Research aimed at quantifying the duration of pain in different intensities under conventional and reformed slaughter methods, as has been done by the Welfare Footprint Project for poultry slaughter, may be helpful in identifying opportunities to reduce time spent in the most intense types of suffering.

Concluding remarks

There are many important caveats to the findings documented in this report. These include uncertainty about the modeling and converting estimates into $/DALY space discussed, the prospective rather than retrospective nature of the analysis, the limited types of cost I included, and the generalizability of the results to other fish welfare interventions in other settings.

Notwithstanding these, my hope is that the report helps improve the understanding of the potential cost-effectiveness of securing stunning commitments for farmed fish commonly consumed in the Mediterranean; how this intervention might compare to alternatives; and the reasons why readers may or may not wish to support this intervention. I also hope the framework and code I’ve used for the analysis may be helpful for others to build on.

Acknowledgements

This report is a project of Rethink Priorities–a think tank dedicated to informing decisions made by high-impact organizations and funders across various cause areas. It was written by Sagar Shah. Thanks to Neil Dullaghan, Ben Stevenson and Bob Fischer for helpful feedback on this draft, Adam Papineau for copy editing assistance, and Ren Ryba, Saulius Šimčikas, Elisa Bianco, Miriam Martinez, William McAuliffe and Willem Sleegers for reviewing and providing feedback on earlier versions of the model and code and results. Thanks also to the 23 stakeholders I interviewed to help build my understanding of the fish farming landscape and inform the scope of the project.

If you are interested in RP’s work, please visit our research database and subscribe to our newsletter.

Annex 1 - Assumptions and data used to estimate unstunned fish consumed annually and slaughter duration by country

Annual consumption by species and country

Annual consumption (in live weight equivalent) for each species in each country was obtained from case study reports produced by the European Market Observatory for Fisheries and Aquaculture Products (EUMOFA) and are described in Table A1.1. Their estimates are based on farmed production data plus imports less exports, with adjustments made to convert trade data related to weights of processed fish products into live weight equivalent. These assumptions feature deterministically in the model. I did not make explicit adjustments for growth in consumption over time, so effectively assume annual consumption remains static over time.³⁵

Table A1.1: Annual consumption of target farmed species (in metric tonnes, live weight equivalent)

Species	Country			Year of estimate
Species	Italy	France	Spain	Year of estimate
Gilthead Seabream	39,449	13,320	18,743	2019
European Seabass	31,122	9,640	25,290	2016
Small Rainbow Trout	24,839	19,816	7,940	2020

Data sources: EUMOFA case studies on Gilthead Seabream, European Seabass and Large Trout.³⁶

Typical slaughter weight by species

Table A1.2 describes the estimated mean fish weights used to translate aggregate consumption by weight into the number of individual fish slaughtered. I used the weight ranges provided by Mood and Brooke (2019a) available on the fishcount.org.uk website (using their small sized estimates for Rainbow Trout). Fish weights entered into the model using a gamma distribution with parameters selected such that approximately 90% of observations fall between the lower and upper estimated mean weights.³⁷

Table A1.2: Estimated mean weight of fish species at slaughter

Species	Weight (grams)		Modeled distribution
Species	Lower	Upper	Modeled distribution
Gilthead Seabream	300	400	Gamma
European Seabass	400	500
Small Rainbow Trout	210	1200

Data source: Mood & Brooke (2019a)

Share of fish stunned before slaughter

Table A1.3 describes assumptions I used for the share of farmed fish consumed in the countries of interest that are stunned before slaughter.

Table A1.3: Share of fish not stunned before slaughter by species

Species	Share stunned		Modeled distribution
Species	Lower	Upper	Modeled distribution
Gilthead Seabream	15%	30%	Uniform
European Seabass	10%	25%
Small Rainbow Trout	20%	50%

For Gilthead Seabream and European Seabass, my assumptions are calibrated using data from the EU Commission Impact Assessment report leaked in June 2023, a recent commitment by a major producer to electrically stun these species, and a survey of Turkish producers.

The leaked EU Commission draft suggested less than 5% of EU production is stunned before slaughter, but does not provide details of the underlying data source or methodology. Consistent with this, a survey of Italian fish slaughter facilities run in 2022 and 2023 found that none of the 21 facilities slaughtering these species used electrical stunning (Clemente at al., 2023).

In December 2023, Avramar, a major producer of these species, committed to implement electrical stunning in all its farms in Spain and more than 50% of its production in Greece by the end of 2027. Avramar reported production of 70,000 tonnes in 2021 (mostly of these species). Given EU consumption of these species was estimated to be just over 240K tonnes in 2021, this commitment alone could plausibly increase the share of Seabass and Seabream consumed in the EU that are electrically stunned by 10 to 20 percentage points.

EUMOFA supply balance sheet data suggest that extra-EU imports accounted for between 20% and 30% of EU consumption of Seabream and Seabass between 2016 and 2021. Turkey is the country with the highest production of these species globally, and is a major exporter to the EU.

A Future for Fish report suggests producers accounting for 76% of Turkish production of Seabream and Seabass indicated that around 40% of their production is electrically stunned before slaughter. If other Turkish producers stun between 0% and 40% of their production, it would suggest between 30% and 40% of Turkish production of these species is electrically stunned.

Taking into consideration the leaked EU report, the Avramar commitment, the Turkish production survey results, the fact that electrically stunned Turkish production is more likely to be exported to the UK or Netherlands than France, Italy, or Spain, and differences between Seabass and Seabream in Turkey – I assumed that between 15% and 30% of Seabream, and 10% to 25% of consumption in the countries of interest will be electrically stunned.

For small Rainbow Trout, I assumed that 20% to 50% of individual fish are stunned before slaughter. In doing so, I assumed that the share of trout farms reported to stun fish in the leaked EU draft impact assessment is a reasonable proxy for the number of individual fish that are stunned.

I modeled uncertainty in the share stunned using a uniform distribution with a minimum reflecting the lower estimate and a maximum reflecting the upper estimate.

Slaughter duration

Table A1.4 describes assumptions for slaughter duration for farmed fish of each species that are not stunned.

Table A1.4: Slaughter duration for fish that are not stunned before slaughter

Species	Slaughter duration (minutes)		Modeled distribution
Species	Lower	Upper	Modeled distribution
Gilthead Seabream	5	40	Gamma
European Seabass	5	40
Small Rainbow Trout	2	10

For Seabream and Seabass, my subjective confidence intervals were based on a Compassion in World Farming Food Business report which identified a range of 5 to 40 minutes across a range of studies.

Finding a reliable slaughter duration for small Rainbow Trout proved more challenging, given the diversity in commonly used slaughter methods. A 2017 EU Commission report suggested that other than electrical stunning, asphyxia in ice is the most common slaughter method for Rainbow Trout, but that immersion in CO₂saturated water is also used in France. Clemente et al (2023) surveyed slaughter facilities in Italy and found 4 of 29 respondents also used asphyxia in air.

Kestin et al. (1991) reported an average duration of 9.6 minutes (with a standard deviation of 1 minute) for small Rainbow Trout to lose visual evoked responses (VERS) when killed by anoxia in air at a temperature of 2°C, aimed at simulating death while packed on ice. The same study found an average duration of 2.6 minutes to VERS at 20°C. I decided to set my subjective confidence interval using these two point estimates, as durations for consciousness persisting CO₂in water reported in a Compassion in World Farming Food Business paper on humane trout slaughter appeared to be within this range.

Annex 2 - The potential cost-effectiveness of non-slaughter interventions

The main body of this report focussed on the potential cost-effectiveness of securing stunning before slaughter commitments from retailers in France, Italy, and Spain. It finds that the cost-effectiveness of this intervention is limited by the relatively short duration of conventional slaughter methods.

The typical lifespan of affected farmed species (1-2 years) is four to five orders of magnitude longer than slaughter duration (well below an hour). This means that non-slaughter interventions providing welfare improvements over a sizable fraction of a fish’s lifespan have the potential to generate larger total welfare gains than humane slaughter interventions, even if the welfare gain at any moment in time is orders of magnitude lower for the non-slaughter intervention.³⁸

This annex explores the potential cost-effectiveness of non-slaughter interventions aimed at improving fish welfare during other stages of the fish life, and the extent to which they might be competitive with selected $/DALY benchmarks.

The calculations in this annex are considerably more speculative than those within the main body. This reflects how corporate asks for non-slaughter interventions for European Seabass and Gilthead Seabream are at a much earlier stage of development than for stunning interventions.³⁹ This in turn reflects gaps in the scientific literature,⁴⁰ heterogeneity in rearing environments,⁴¹ difficulty in achieving consensus over the highest priority asks,⁴² and a desire to ensure any specific measures are feasible (for example, through economic evaluation and a successful pilot program) before demanding them.

In the time I worked on this project, I was unable to produce specific estimates of the duration of benefits, magnitudes of welfare benefits, and probability of success of a range of non-slaughter welfare asks. To illustrate the potential cost-effectiveness of non-slaughter interventions, I instead produced estimates conditional on hypothetical values about the fraction of a fish’s lifespan affected by an intervention and how meaningful intervention might be. As with the slaughter intervention calculations, improvements are either expressed as a % of the fish welfare range, or the moral value of gaining the benefits of the intervention for a fish life year relative to gaining a year of human life at full health.

Assumptions and method

Table A2.1 – Assumptions for non-slaughter intervention cost-effectiveness calculations

Assumption description		Value or 90% Subjective Confidence Interval	Distribution
Costs per country (same for all countries)		EUR400K	n/a
Probability of success (same for all countries)		5% to 15%	n/a
Market share of retailers making commitments	France	0.1% to 24%	Beta
	Spain	0.1% to 14%	Beta
	Italy	0.1% to 10%	Beta
Typical fish lifespan	Small Rainbow Trout	12 to 15 months	Gamma
	European Seabass	14 to 24 months	Gamma
	Gilthead Seabream	12 to 20 months	Gamma

Table A2.1 describes assumptions used for the non-slaughter welfare intervention cost-effectiveness estimates, where they differ from those described in the Methods section and Annex 1. To simplify the exposition of results, I chose to use only a single scenario. I assumed higher costs per country than both slaughter intervention scenarios, reflecting the higher potential costs of establishing priority asks, building consensus and running pilot programs. I also assumed a lower probability of success, given interventions that might have the highest potential to improve welfare might be more difficult to implement or have higher economic costs.

Reflecting an assumption that corporate commitments for non-slaughter interventions will typically be run after some retailers will have already made stunning commitments, I selected market share assumptions that reflect the lower bound of the pilot stunning scenario assumption, and upper bound of the scale stunning scenario assumption.

The calculations closely largely follow the approach described in the methodology section. The main difference is that I first calculated the lifespan of fish affected per dollar (by modifying the calculations described in Figures 4 to 6), and then calculated $/DALY by inversing the product of the total life span of fish per dollar, the share of a fish’s lifespan affected by an intervention and the DALY conversion factor (a slight modification to Figure 7).⁴³ Code and calculations, as well as additional charts and plots, are available from GitHub.

Results

Fish affected per dollar and their associated lifespan

Table A2.2 describes the estimates of the fish affected per dollar and the total lifespan of these fish in years, given the modeling assumptions. The estimated number of fish affected per dollar for welfare interventions is lower than for stunning commitments, but their collective lifespan is thousands of times longer than the aggregate slaughter duration. That said, the collective lifespan of fish affected by welfare interventions per dollar spent is significantly lower than estimates of the number of chicken years affected per dollar from historical chicken campaigns (10 to 120).

Table A2.2 – Assumptions for non-slaughter intervention cost-effectiveness calculations

Measure	Statistic
	Mean	Percentile
	Mean	5th	95th
Fish affected per dollar	0.7	0.1	2.4
Lifespan of fish affected per dollar (in years)	1.0	0.1	3.1

$/DALY for a hypothetical non-slaughter intervention affecting 10% to 25% of a fish’s lifespan

Figure A2.1 provides a density plot of the $/DALY distribution for a hypothetical welfare intervention that provides benefits to a fish lasting between 10% and 25% of its lifespan, and provides average welfare gain between 1% and 10% of the fish’s entire welfare range (negative to positive) over this period.⁴⁴

Ninety per cent of simulations produce an estimate lying between $88.6 and $1.23M per DALY averted, and a median estimate of $2.41K. The cost of averting a DALY equivalent for the hypothetical welfare intervention is considerably lower than for stunning intervention in the scale scenario (5th percentile: $10.4K; 95th percentile $114M; median: $216K).

For the hypothetical welfare intervention, approximately 2% of simulations beat a benchmark of $50/DALY, 36% beat $1K/DALY, and 85% beat $70K/DALY. This might suggest that an intervention generating such impacts might not look competitive with the most promising human global health and development interventions, but could potentially be perceived as comparable to general health expenditure in low HDI countries.

Figure A2.1 – Cost-effectiveness of non-slaughter intervention affecting 10% to 25% of a fish’s lifespan – $/DALY density plot

Share of simulations beating a $50/DALY benchmark – welfare range and moral value approaches

Figures A2.2 and A2.3 illustrate the share of simulations beating a benchmark of $50/DALY benchmark given different assumptions about the fraction of a fish’s life affected by the intervention. Figure A2.2 provides results calculated using the welfare range approach – conditional on the size of the welfare gain arising from the intervention, expressed as a share of the fish welfare range. Figure A2.3 provides the share conditional on assumptions about the moral value of improving a fish life year through the intervention relative to averting a DALY. Tables of results featuring the share of simulations beating other benchmarks are available from GitHub.

Using the welfare range approach, an intervention that affects half of a fish’s entire lifespan would need to generate an average welfare gain equivalent to around three quarters of a fish’s entire welfare range in order for 50% of simulations to beat a benchmark of $50/DALY. It would probably be very difficult to find a welfare intervention or set of interventions that achieves such benefits in reality. Half a fish’s lifespan is a very long amount of time to affect,⁴⁵ and a three quarters increase in the fish welfare range is very large, given the range runs from the worst possible to the best possible valenced experience.

That said, it is possible that specific modeling assumptions could materially change how cost-competitive the intervention looks compared to benchmarks. For example, if I instead assumed the welfare range of affected farmed fish relative to humans was double the placeholder estimate for salmon, and that the welfare improvement of 10% of the human welfare range (instead of 50%) for a year was equivalent to averting a DALY, then an intervention improving fish welfare by 10% of its welfare range lasting a quarter of its lifespan could beat the $50/DALY benchmark in 43.8% of simulations.⁴⁶

Figure A2.2 – Share of simulations beating $50/DALY – welfare range approach

Under the moral value approach, an intervention that provided benefits for 10% of a fish’s lifespan could beat a $50/DALY benchmark in 64.3% of simulations, for someone who viewed improving 2 years of fish life via the intervention as morally equivalent to gaining an additional year of human health at full health.

Figure A2.3 – Share of simulations beating $50/DALY – moral value approach

Discussion

The results of this annex suggest that, conditional on the modeling assumptions and estimation framework used, non-slaughter welfare interventions affecting a sizable fraction of a fish’s lifespan appear to have the potential to be considerably more cost-effective than stunning interventions. Although they do not look competitive with the most promising health and development interventions using my baseline assumptions for translating results into $/DALY space, it’s possible they could look competitive with alternative assumptions which I also consider reasonable.

Given that corporate asks for non-slaughter welfare interventions are still at a relatively early stage of development for the farmed species most commonly consumed in France, Italy, and Spain, the results remain speculative and theoretical. Within the costs I have assumed, it is unclear whether it will be possible for NGOs to reach consensus on interventions that industry could plausibly implement that will generate meaningful improvements over a sizable fraction of a fish’s life, as well as run necessary pilots and secure commitments.

Nevertheless, given the theoretical potential cost-effectiveness of longer-duration non-slaughter interventions, I think there would be value in NGOs who plan to secure stunning commitments from retailers to use this engagement as an opportunity to build momentum for future welfare interventions, for example, by getting retailers to recognise fish sentience and the importance of fish welfare.

References

Animal Ask (2023). Farmed fish advocacy in Denmark. https://perma.cc/KF3U-VY6M

Berger, A. & Oehlsen, E. (2023). Our planned allocation to GiveWell’s recommendations for the next few years. Open Philanthropy Blog Post. https://perma.cc/N9JF-G7QW

Bollard, L. (2016). Initial Grants to Support Corporate Cage-free Reforms. Open Philanthropy Blog Post. https://perma.cc/N26B-JZ93

Buonanno (2022) Probability distributions of Cost-Effectiveness can be misleading. Effective Altruism Forum. https://perma.cc/4X2B-HQQS

Capriati, M. (2018). Cause Area Report: Corporate Campaigns for Animal Welfare. Founders Pledge. https://perma.cc/6REJ-ZN6K

Clemente, G. A., Tolini, C., Boscarino, A., Lorenzi, V., Dal Lago, T. L., Benedetti, D., Bellucci, F., Manfrin, A., Trocino, A., & Rota Nodari, S. (2023). Farmed fish welfare during slaughter in Italy: survey on stunning and killing methods and indicators of unconsciousness. Frontiers in Veterinary Science, 10(1253151). https://doi.org/10.3389/fvets.2023.1253151

Compassion in Food Business. (n.d.) Humane Slaughter: Rainbow trout https://perma.cc/8TF3-PDCC

Compassion in Food Business (n.d.) Humane Slaughter: Gilthead sea bream and European sea bass https://perma.cc/PGA2-JBMU

Daroudi, R., Akbari Sari, A., Nahvijou, A. et al. (2021) Cost per DALY averted in low, middle- and high-income countries: evidence from the global burden of disease study to estimate the cost-effectiveness thresholds. Cost Effectiveness and Resource Allocation 19(7) . https://doi.org/10.1186/s12962-021-00260-0

Duffy, L. (2022). Welfare Range and Probability of Sentience Distributions. Rethink Priorities. https://perma.cc/UYD6-J6XX

Dullaghan, N. (2023). EU farmed fish policy reform roadmap brief. Rethink Priorities. https://perma.cc/22VU-ZWGB

EdoArad (2023) ‘1/E(X) is not E(1/X)’. Effective Altruism Forum. https://perma.cc/K2TU-3AEC

EFSA Panel on Animal Health and Welfare (2009). Species-specific welfare aspects of the main systems of stunning and killing of farmed fish: Rainbow Trout. EFSA Journal, 7(4):1012, 55 pp. doi:10.2903/j.efsa.2009.1012

EUMOFA (2016). Case study. Price structure in the supply chain for fresh carp in Central Europe. European Market Observatory for Fisheries and Aquaculture Products. https://perma.cc/G9K7-WY3X

EUMOFA (2019). Case study. Seabass in the EU. Price Structure in the Supply Chain for Seabass. European Market Observatory for Fisheries and Aquaculture Products. https://perma.cc/9AT9-2W26

EUMOFA (2021). Case study. Portion trout in the EU. Price Structure in the Supply Chain. European Market Observatory for Fisheries and Aquaculture Products. https://perma.cc/G47F-CS6P

EUMOFA (2021). Freshwater Aquaculture in the EU. European Market Observatory for Fisheries and Aquaculture Products. https://perma.cc/3WCU-EHSY

EUMOFA (2019). Case study. Fresh Gilthead Seabream in the EU. Price Structure in the Supply Chain. European Market Observatory for Fisheries and Aquaculture Products. https://perma.cc/VQ2G-HDP5

EUMOFA (2022). The EU Fish Market 2022 Edition. European Market Observatory for Fisheries and Aquaculture Products. https://perma.cc/TQF4-B5N2

EUMOFA (2023). Case study. Large trout in the EU. Price Structure in the Supply Chain. European Market Observatory for Fisheries and Aquaculture Products. https://perma.cc/XV5N-6B5R

European Commission (2017). Welfare of farmed fish: Common practices during transport and at slaughter. https://perma.cc/LV9N-QAH7

European Commission (2023). Leaked Impact Assessment Report and Annex. [Document cannot be shared under terms under which we received it].

Fischer, B. (2022). An Introduction to the Moral Weights Project. Rethink Priorities. https://perma.cc/MJ3V-GMGX

Fischer, B. (2022). Rethink Priorities’ Welfare Range Estimates. https://perma.cc/4BNL-FUHB

Fischer, B. & Clatterbuck, H. (2023). Causes and uncertainty: Rethinking value in expectation. Rethink Priorities. https://perma.cc/UKV3-MN28

Gaffney, L. P., & Lavery, J. M. (2022). Research Before Policy: Identifying Gaps in Salmonid Welfare Research That Require Further Study to Inform Evidence-Based Aquaculture Guidelines in Canada. Frontiers in Veterinary Science, 8. https://doi.org/10.3389/fvets.2021.768558

Jokumsen, A., & Svendsen, L. M. (2010). Farming of freshwater rainbow trout in Denmark. DTU Aqua. DTU Aqua-rapport No. 219-2010. https://perma.cc/T3NM-FKRL

Kıraç Uncu, D. & Özen, D. (2023) ‘Electrical Stunning System: Türkiye Review’. Future for Fish. https://perma.cc/6NBL-JEPB

Kestin, S & Wotton, Stephen & Gregory, N. (1991). Effect of slaughter by removal from water on visual evoked activity in the brain and reflex movement of rainbow trout (Oncorhynchys mykiss). The Veterinary Record. 128. 443-6. 10.1136/vr.128.19.443. https://perma.cc/GDG7-NTYL

McAuliffe W., & Shriver, A. (2023). “Dimensions of Pain” Workshop: Summary and Updated Conclusions. Rethink Priorities. https://perma.cc/H24Y-7G2E

McAuliffe W, Shriver A. (2022) The Relative Importance of the Severity and Duration of Pain. Open Science Framework Preprints. https://osf.io/ezvr2/

Mood, A. & Brooke, P. (2019). Estimated numbers of farmed fishes living in global aquaculture. Spreadsheet. https://perma.cc/VB99-T48E

Mood, A. & Brooke, P. (2019). Estimated numbers of individuals in global aquaculture production (FAO) of fish species (2017). Spreadsheet. https://perma.cc/M3LC-VL8D

Mood, A., Lara, E., Boyland, N., & Brooke, P. (2023). Estimating global numbers of farmed fishes killed for food annually from 1990 to 2019. Animal Welfare, 32(E12). https://doi.org/10.1017/awf.2023.4

Norwegian Veterinary Institute. (2023). ‘Norwegian Fish Health Report 2022’ https://perma.cc/FZF4-GYPU

Oehlsen, E. (2023). Comment on “Open Phil Should Allocate Most Neartermist Funding to Animal Animal Welfare”. Effective Altruism Forum. https://perma.cc/6WZ2-HVLJ

Pavlidis, M., Papaharisis, L., Adamek, M., Steinhagen, D., Jung-Schroers, V., Kristiansen, T., Theodoridi, A., Otero Lourido, F. (2023). Research for PECH Committee – Animal welfare of farmed fish, European Parliament, Policy Department for Structural and Cohesion Policies, Brussels https://perma.cc/SN6T-K4WB

Ryba, R., & van Pelt, K., (2023) ‘Slaughter of farmed seabass and seabream in Europe: Literature review’ Effective Altruism Forum. https://perma.cc/L46N-F6GW

Saraiva, J. L., Rachinas-Lopes, P., & Arechavala-Lopez, P. (2022). Finding the “golden stocking density”: A balance between fish welfare and farmers’ perspectives. Frontiers in Veterinary Science, 9. https://doi.org/10.3389/fvets.2022.930221

‌

Saraiva, J. L., Faccenda, F., Cabrera-Álvarez, M. J., Povinelli, M., Hubbard, P. C., Cerqueira, M., Farinha, A. P., Secci, G., Tignani, M. V., Pulido Rodrigeuz, L. F., & Parisi, G. (2023). ‘Welfare of rainbow trout at slaughter: Integrating behavioural, physiological, proteomic and quality indicators and testing a novel fast-chill stunning method’. Aquaculture, 581(740443). https://doi.org/10.1016/j.aquaculture.2023.740443

Scientific Opinion of the Panel on Animal Health and Welfare on a request from the European Commission on welfare aspect of the main systems of stunning and killing of farmed seabass and seabream. The EFSA Journal (2009) 1010, 1–52. https://doi.org/10.2903/j.efsa.2009.1010

Schuck-Pain, C., Alonso, W., & Hamilton, C. (2024). ‘Short agony or long ache: comparing sources of suffering that differ in duration and intensity’. Welfare Footprint Project, Center for Welfare Metrics. Forthcoming.

Šimčikas, S. (2019). List of ways in which cost-effectiveness estimates can be misleading. Effective Altruism Forum. https://perma.cc/9BLQ-WP2V

Šimčikas, S. (2019). Corporate campaigns affect 9 to 120 years of chicken life per dollar spent. Rethink Priorities. https://perma.cc/U9D2-AJ4D

Šimčikas, S. (2019). Estimates of global captive vertebrate numbers. Rethink Priorities. https://perma.cc/M7RQ-KDC5

Shrimp Welfare Project (2023). Shrimpact. https://perma.cc/FMS7-BNH9.

Springlea, R. (2022). Economic evaluation of humane slaughter methods for farmed fish In Greece. Essere Animali & Animal Ask. https://perma.cc/J8LS-QTMW

Springlea, R. (2022). Economic evaluation of humane slaughter methods for farmed fish In Italy. Essere Animali & Animal Ask. https://perma.cc/EDX9-PZW9

World Organization for Animal Health. (2023). “Welfare aspects of stunning and killing of farmed fish for human consumption”, Aquatic Animal Health Code 1–4. https://perma.cc/XD4S-SNP4

Notes

Though there are plans in place to investigate this for Atlantic Salmon. ↩
For example, an Animal Charity Evaluators report first published in 2019 reported they were aware of only four charities that had completed undercover investigations of fish farms, and a database tracking undercover fish investigations only includes two investigations released prior to 2017. Staff at Animal Charity Evaluators rated farmed fish as the most neglected among all vertebrates in their animal prioritization framework. ↩
Since 2016, around a tenth of the farm animal welfare grants recommended by Open Philanthropy by value have referenced aquatic animals in their grant title, according to their grant database (as of November 22, 2023) ↩
For example, the Fish Welfare Initiative. ↩
For example, the Aquatic Animal Alliance, run by the Aquatic Life Institute, lists over 130 members as of November 22, 2023. ↩
For example see the Fish Welfare Initiative cost-effectiveness ratings here. ↩
As of November 23, 2023, Chicken Watch listed 374 broiler commitments and 1425 cage-free commitments in Europe. Compare this to 238 broiler commitments and 596 cage-free commitments in North America, the region with the second-highest number of commitments. Most of the top 10 grocers in France have signed up to the European Chicken Commitment. ↩
See broiler commitments listed on Chicken Watch, and European Supermarket report listings of the largest supermarket chains in France and the Netherlands. ↩
According to this report, there are bans in place in Austria, Luxembourg, and Switzerland, with enacted bans due to come into force in Germany and the Czech Republic. A ban has also been agreed on in Denmark. ↩
Animal Ask (2023) includes a list of German grocers who have made public commitments to sell only third-party certified farmed fish, including Edeka, Lidl, Kaufland, Rewe, Aldi Nord, and Metro. ↩
For example Tesco, Sainsbury’s, Morrisons & Waitrose report that 100% of their own brands of farmed salmon, trout, seabream, seabass, and pangasius are stunned before slaughter, while Marks & Spencer reports that all its farmed fish are third-party certified. ↩
For example, Hilton Foods, a food supplier, set itself an objective of 100% stunned farmed fish in 2014, and achieved 100% electrical stunning for Seabass and Seabream by 2018 (see page 16). ↩
Based on EUMOFA supply balance sheet data, which suggests that in 2021 apparent consumption of farmed finfish was around 1.9M tonnes, compared to production of 0.6M, imports of 1.6M and exports of 0.2M. ↩
Fish estimated mean weights mostly obtained from Mood & Brooke (2019a and 2019b). For Atlantic Salmon, mean weight imputed from the 2022 Norwegian Fish Health Report. Mean weight of 500g assumed for small Rainbow Trout. Distribution of consumption between small and large Rainbow Trout based on apparent consumption estimates in EUMOFA case study reports on portion and large Rainbow Trout. ↩
Norway is the world’s largest farmed salmon producer, and the biggest exporter to the EU. Norwegian legislation requires farmed fish to be stunned prior to slaughter. ↩
Country-level estimates of small Rainbow Trout apparent consumption based on the residual of estimates of total Rainbow Trout consumption and large Rainbow Trout consumption from this EUMOFA report. ↩
See this talk (from 20:11) by Claudia Millán Caravaca, which was presented shortly before Equalia rebranded to AWO. ↩
For example, in Spain, the market share for the pilot scenario was calibrated around the market share of two regional grocers who had signed up to the European Chicken Commitment, and in Italy, around the market share of the Italian subsidiary of Metro AG, a german grocer that has made commitments to sell certified fish in Germany and to observe stocking density limits on fish sold in Turkey. ↩
See Arad (2023) and Buonanno (2022). ↩
For France and Spain, I used Kantar World Panel data to calibrate market shares, but I calculated them manually for Italy using data from this European Supermarket Magazine article. I excluded Mercadona from the Spanish grocery shares on account of it likely being a higher priority for a broiler commitment. ↩
The assumption feeds through linearly into the number of fish affected per dollar, so halving this assumption would halve the number of fish affected per dollar. ↩
Hilton Foods, a meat and seafood supplier, set a target to switch to 100% of fish stunned before slaughter in 2014, and achieved 100% electrical stunning for Seabass and Seabream by 2018. An Aquaculture Stewardship Council consultation on fish welfare standards in 2022 suggested a transition time of 3 years for stunning of Seabass and Seabream before slaughter. The consultation feedback report did not highlight major concerns with the criterion. ↩
At the time this project started, it seemed possible the EU might include a fish slaughter policy as part of a broader animal welfare reform package. A draft EU Commission impact assessment document leaked in June 2023 suggested farmed fish slaughter reforms would have a 10-year implementation period for Gilthead Seabream and European Seabass. ↩
Like Šimčikas (2019), my cost assumptions do not include future spending on actions to ensure retailers comply with any commitments they make. ↩
See Chart 27 on page 54. My understanding is that sales of the farmed species of interest would be classified as unprocessed (i.e., sold whole, gutted, or fileted) rather than processed (e.g., sold in glass jars, canned, in oil or brine, or with sauce). ↩
For example, the 2022 EUMOFA case study on Seabream suggests supermarkets and hypermarkets accounted for around three quarters of retail sales. ↩
This is a reasonable assumption for Rainbow Trout, which is closely related to salmon and a member of the salmonid family. The choice for Seabream and Seabass is less obvious, but salmon seemed more relevant than carp (the other finfish for which the Rethink Priorities Moral Weight Project produced estimates) given that salmon is carnivorous and has been cultured for a similar amount of time as Seabream and Seabass. ↩
There are good reasons to believe a much smaller increase in welfare for a year should be equivalent to averting a DALY. For example, if one believes the best human welfare state far exceeds the average welfare of the average human in full health, or if one believes the worst valenced experiences might be considerably more intense than the best ones. Changing the assumption so that averting a DALY would be equivalent to welfare improvement of 10% of the human welfare range for a year, would result in a fivefold reduction in the $/DALY estimates for the stunning intervention, making it appear more competitive. But it would not materially change the conclusions of how competitive the intervention looks when compared to alternatives. ↩
In a comment on the Effective Altruism Forum in 2023, Emily Oehlsen, Managing Director of Open Philanthropy, a major funder of chicken corporate campaigns, reported that since 2016, “we’ve covered many of the strongest opportunities in this space, and we think that current marginal opportunities are considerably weaker,” and that the marginal farm animal welfare funding opportunity might be a fifth as cost-effective as the average from Šimčikas (2019). Šimčikas also made a comment in 2021 about unpublished estimates of chicken welfare reforms, suggesting cost-effectiveness was 2 to 3 times lower in 2019-2020 than in 2016-2018. Similar sentiments are discussed here and here. ↩
If the individual simulations from Šimčikas (2019) affected 3.5 times to 5 times fewer chickens per dollar, the distribution would look very similar to the fish stunning scale scenario, and would produce a higher (or lower) number in about half of simulations. See GitHub for these results. ↩
For example, for cage-free layer hen campaigns, the Rethink Priorities Cross-Cause Model (write-up, tool) generates an average cost per DALY averted of $1.40 across all its simulations under the default parameters in February 2024. ↩
Note that welfare gain assumption here refers to the average improvement in welfare across the entire duration of slaughter without stunning, rather than being restricted to 50-90% of the duration. ↩
The headline findings from the Annex on non-slaughter interventions do not currently imply that such interventions will necessarily be competitive with the best animal interventions. But that analysis assumed a probability of success of only 5% to 15%. If it turns out that retailers in these countries seem receptive to fish welfare commitments, it might be reasonable to double the probability of success. When coupled with alternative but reasonable assumptions about improvements in the human welfare range relative to averting a DALY, it seems plausible that such interventions could clear a hypothetical cost-effectiveness bar for animal welfare interventions. ↩
An exploratory analysis I conducted suggested it seems plausible that farmed fish stunning interventions in these countries could avert more time in excruciating pain than historical chicken campaigns, though it seems unlikely they could avert anywhere near as much time in less intense types of pain. See GitHub for this analysis. ↩
I did not include an explicit adjustment to keep the model simple and because I do not believe it will have a material impact on final results. EUMOFA historical balance sheet data for the EU28 suggest a consumption growth rate of 3 to 4% for Gilthead Seabream and European Seabass from 2008-2019, and of -0.5% for Rainbow Trout. One way to adjust for the impact of consumption growth in the current model set up is to modify the years of impact figure. ↩
Small Rainbow Trout estimates calculated from the residual of total trout and large trout consumption, using Associazione Piscicoltori Italiani estimates for large trout consumption (see pages 32 to 34 for the EUMOFA Large Trout report). ↩
Note that during conversations with NGOs operating in Spain and Italy, I heard that there was a tendency towards selling Seabream and Seabass at heavier weights in recent years. Due to the timing of when I received this information, I did not make adjustments to take this into account. I do not think adjusting the numbers would materially change the main conclusions of the results as it would reduce the number of fish affected. ↩
At least, in estimation frameworks which multiply duration by intensity. ↩
For example, Essere Animali’s template for corporate fish welfare policies includes stunning before slaughter, responsible use of antibiotics, an articulation of the company’s position on genetic engineering (breeding of genetically modified, cloned or triploid animals should not be permitted), and routine staff training as essential elements, while stocking density, water quality, maximum time out of water, environmental enrichments and the use animal welfare indicators are listed as recommendations. Essential elements are described as fundamental aspects for fish welfare, while recommended elements are described as including more cutting-edge aspects of fish welfare that might help a company demonstrate vision and leadership.
At the time of writing, AWO’s corporate ask was a stunning commitment, but also required companies to recognise fish as sentient beings.
Corporate asks secured by Future for Fish have included stocking density limits and a requirement for production to be third-party certified, as well as an electrical stunning requirement. ↩
There is considerably more academic research on salmonid welfare than there is for Mediterranean species like European Seabream and Gilthead Seabass. Yet, even in the context of developing evidence-based aquaculture guidelines in Canada, Gaffney & Lavery (2022) find several areas of salmonid welfare (including biodensity, sea lice prevention, and enclosure design) where scientific evidence remains lacking. Pavlidis et al. (2023) highlight a need for further research for juvenile life stages for European Seabream and Gilthead Seabass. ↩
For example, see page 35 of Pavlidis et al. (2023). ↩
Stocking density limits are an example of a potential welfare ask that could affect welfare for a large fraction of a fish lifespan and have the desirable feature of being relatively straightforward to describe and measure, at least among fish welfare measures. Indeed they already feature in some commitments. But Saraiva et al. (2022), authored by prominent fish welfare researchers, argues that seeking to directly limit stocking density may in fact be unworkable and ineffective. During interviews I conducted as part of this project, several respondents noted the difficulty in identifying consensus in non-slaughter welfare interventions. ↩
Note that in the calculations, I continued to use the share of fish currently stunned before slaughter as a proxy for the number of fish who might not benefit from a given non-slaughter welfare commitment. ↩
Both distributions were modeled using a beta distribution. ↩
It is, however, theoretically possible that an intervention that improves fish health or welfare at the juvenile phase could provide benefits that endure throughout a fish’s life. ↩
I think these are both reasonable assumptions given the Rethink Priorities Moral Weights Project’s method for estimating welfare ranges assigned lower scores to less-researched species. I also think it’s plausible that the difference between the ‘neutral point’ and the average valenced experience for the average human in full health occupies a relatively small share of the human welfare range. ↩

Changelog

[March 18, 2024] - Table 13 and figures 13 & A2.3 and associated descriptions corrected. Original text referred to “gaining to a human DALY”, when in fact they should have referred to “averting a human DALY”.

Prospective cost-effectiveness of farmed fish stunning corporate commitments in Europe

Executive summary

Caveats and limitations

Definitions

Introduction and motivation

Methodology

Estimating the number of unstunned fish consumed annually and slaughter duration by country

Estimating the number of fish affected and fish hours affected per dollar

Commitment scenarios - market share and success probabilities

Other assumptions – years of impact, implementation discounts, and grocery sales share

Converting the results into $/DALY averted equivalents

Results

Number of fish affected

Number of fish hours affected

Cost-effectiveness in $ per DALY averted

Share of simulations beating $/DALY benchmarks under the welfare range approach

Share of simulations beating $/DALY benchmarks under the moral value approach

Summary and discussion

Summary of results

Theoretical potential for non-slaughter interventions to look more effective

Possible reasons to support fish stunning corporate commitment efforts

Concluding remarks

Acknowledgements

Annex 1 - Assumptions and data used to estimate unstunned fish consumed annually and slaughter duration by country

Annual consumption by species and country

Typical slaughter weight by species

Share of fish stunned before slaughter

Slaughter duration

Annex 2 - The potential cost-effectiveness of non-slaughter interventions

Assumptions and method

Results

Fish affected per dollar and their associated lifespan

$/DALY for a hypothetical non-slaughter intervention affecting 10% to 25% of a fish’s lifespan

Discussion

References

Notes

Changelog

About

Our Work

Prospective cost-effectiveness of farmed fish stunning corporate commitments in Europe

Executive summary

Caveats and limitations

Definitions

Introduction and motivation

Methodology

Estimating the number of unstunned fish consumed annually and slaughter duration by country

Estimating the number of fish affected and fish hours affected per dollar

Commitment scenarios - market share and success probabilities

Other assumptions – years of impact, implementation discounts, and grocery sales share

Converting the results into $/DALY averted equivalents

Results

Number of fish affected

Number of fish hours affected

Cost-effectiveness in $ per DALY averted

Share of simulations beating $/DALY benchmarks under the welfare range approach

Share of simulations beating $/DALY benchmarks under the moral value approach

Summary and discussion

Summary of results

Theoretical potential for non-slaughter interventions to look more effective

Possible reasons to support fish stunning corporate commitment efforts

Concluding remarks

Acknowledgements

Annex 1 - Assumptions and data used to estimate unstunned fish consumed annually and slaughter duration by country

Annual consumption by species and country

Typical slaughter weight by species

Share of fish stunned before slaughter

Slaughter duration

Annex 2 - The potential cost-effectiveness of non-slaughter interventions

Assumptions and method

Results

Fish affected per dollar and their associated lifespan

$/DALY for a hypothetical non-slaughter intervention affecting 10% to 25% of a fish’s lifespan

Discussion

References

Notes

Changelog

Research summary: The evolution of nociception in arthropods

Pre-slaughter mortality of farmed shrimp

About

Our Work